Informative Clustering of Microarray Data Using Binary and Survival Outcomes

• Activity Number: 402
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Jessie Hsu*+ and Dianne M Finkelstein and David A Schoenfeld
• Companies: Harvard University and Massachusetts General Hospital/Harvard University and Harvard Medical School
• Keywords: clustering ; microarray ; Bayesian ; MCMC ; data augmentation

Abstract

The rapid technological development of high-throughput genomics has given rise to extremely complex high-dimensional microarray data sets. One strategy for analyzing microarray experiments is to carry out a cluster analysis to find groups of genes that behave similarly. Sometimes the clusters are related to clinical outcomes but this is usually done separately. However, using clinical outcomes to drive the clustering can potentially result in clusters that are more descriptive of the clinical course of disease. We propose a joint hierarchical model that relates gene expression measurements to each other as well as to patient outcome. Gene expression is modeled using normally distributed cluster random effects that are linearly related to outcome via latent continuous variables where the probability or hazard of outcome depends on these latent variables. A Markov chain Monte Carlo procedure is used to sample the model parameters and determine the clustering pattern. We will apply our method to microarray data collected from trauma patients in the Inflammation and Host Response to Injury program.